Institutional Repository of Vilnius University
Meno kūrinių stiliaus klasifikavimo sistemos kūrimas /
Abstract
dc:descriptionThis paper examines the problem of art style classification using deep learning methods and a set of images from the WikiArt data source. The aim of the research is to evaluate how different pre-trained convolutional neural network architectures and their partial adaptations affect artistic style recognition, as well as to evaluate the applicability of the results in a practical prototype context. In the experimental part, a comparison of three architectures was performed, applying feature extraction and partial adaptation to each model. The quality of the models was assessed by accuracy, macro F1 and additional metrics, including Top-k accuracy and macro AUC. The results showed that partial adaptation consistently improves classification performance, but the extent of the improvement depends on the architecture, and the use of fixed representations alone is not always sufficient for reliable style separation. An additional test with an improved configuration showed the possible increase in performance without changing the base network architecture. For practical interpretation, a prototype was created in which the results are presented in a recommendatory manner, showing the user several most likely style alternatives with their probabilities. Such presentation is compatible with the specifics of the task, when some styles have overlapping visual features, therefore presenting several candidates is more informative than a single categorical solution.
Degree
thesis:*- Grantor dc:publisher
- Institutional Repository of Vilnius University
- Year dc:date
- 2026
Author and committee
dc:creator, dc:contributor.*- Author dc:creator
-
- Švilpa, Dominykas,
Rights
dc:rights- Statement dc:rights
-
- info:eu-repo/semantics/openAccess
- Language dc:language
- lit
Identifiers
dc:identifier.*- Repository record dc:identifier
- https://repository.vu.lt/VU:ELABAETD277691390&prefLang=en_US